Page History

...

Yearly Solr sharding is a routine that can drastically improve the performance of your DSpace SOLR statistics. It was introduced in DSpace 3.0 and is not backwards compatible. The routine decreases the load created by the logging of new usage events by reducing the size of the SOLR Core in which new usage data are being logged. By running the script, you effectively split your current SOLR core, containing all of your usage events, into different SOLR cores that each contain the data for one year. In case your DSpace has been logging usage events for less than one year, you will see no notable performance improvements until you run the script after the start of a new year. Both writing new usage events as well as read operations should be more performant over several smaller SOLR Shards instead of one monolithic one.

It is highly recommended that you execute this script once at the start of every year. To ensure this is not forgotten, you can include it in your crontab or other system scheduling software.  Here's an example cron entry (just replace [dspace] with the full path of your DSpace installation):

# At 12:00AM on January 1, "shard" the DSpace Statistics Solr index.  Ensures each year has its own Solr index - this improves performance.
0 0 1 1 * [dspace]/bin/dspace stats-util -s

Technical implementation details

After sharding, the SOLR data cores are located in the [dspace.dir]/solr directory. There is no need to define the location of each individual core in solr.xml because they are automatically retrieved at runtime. This retrieval happens in the static method located in theorg.dspace.statistics.SolrLogger class. These cores are stored in the statisticYearCores list each time a query is made to the solr these cores are added as shards by the addAdditionalSolrYearCores method. The cores share a common configuration copied from your original statistics core. Therefore, no issues should be resulting from subsequent ant updates.

The actual sharding of the of the original solr core into individual cores by year is done in the shardSolrIndex method in the org.dspace.statistics.SolrLogger class. The sharding is done by first running a facet on the time to get the facets split by year. Once we have our years from our logs we query the main solr data server for all information on each year & download these as csv's. When we have all data for one year we upload it to the newly created core of that year by using the update csvhandler. One all data of one year has been uploaded that data is removed from the main solr (by doing it this way if our solr crashes we do not need to start from scratch).